If you are interested in this topic

inoccotepola222

Read now PureFluent launches WordStore – the first subscription for translations Read now How can I as a customer assess the translation quality? Read now Interview with Pentland Brands about optimizing the customer experience on Amazon Read now July 24, 2020 As we all know, machine translation is getting better and better. If you would like to find out a little more about it, just read our post “How good is machine translation?”. In today's post I would like to give some practical tips on how to achieve better machine translations. Some machine translation engines are better than others I need to define this heading a little more precisely.

The translation quality of some machine translation engines, the so-called HK Phone Number MT engines, is better than others for certain language combinations and subject areas. Memsource has found an interesting solution for this. For this purpose, various engines are evaluated with regard to translation quality for specific language combinations and specialist areas. On this basis, the most suitable option for the respective project is then suggested. In summary, this means that THE best engine does not exist. To get the best result, you need to have multiple options. I recommend Memsource's latest report on machine translation. Customer-specific trained engines deliver better translations MT engines for machine translation of your content already deliver surprisingly good results.



But they don’t know your industry, the style you want, or company-specific terminology. If your translation memory is large enough, you can use it to train such an engine for your purposes. To do this, you need at least 10,000 entries, i.e. sentence pairs, for each language combination. But the more, the better. This corpus forms your “training data” and must come from high-quality human translations. Otherwise it’s like, “If crap comes in, crap comes out!” Optimize training data Even if this seems strange at first; What you can provide in terms of training data is initially a bit messy and not really usable in the world of the MT engine.